Spin Torque and Magnetic Tunnel Junctions

Ed Myers, Frank Albert, Ilya Krivorotov, Sergey Kiselev, Nathan Emley, Patrick Braganca, Greg Fuchs, Andrei Garcia,

Ozhan Ozatay, Eric Ryan, Jack Sankey, John Read, Phillip Mather, Dan Ralph

Jordan Katine and Daniele Mauri (HGST)

Outline

Spin torque switching in spin valves Switching speedsAsymmetry of switching currents (spin torque and spin accumulation)Reducing switching current levels

Non-uniform spin torque systemsSwitching by concentrated spin current injectionVortex spin torque oscillator

Spin torque in magnetic tunnel junctionProbing spin torque as function of tunnel junction bias

Realizing Spin Transfer Effects

Nanopillar GMRSPIN VALVE

Py (2 nm)

Py (12 nm)Cu (6 nm)

Cu

Cu

free layer

fixed layer

Conventional ferromagnet spin transfer devices require lateral dimensions ≤250 nm to avoid significant self-field effects from required current levels

Low impedance ~ 0.01 Ω-µm2

GMR (∆R/R) ~ 10-20%High impedance ~ 1 - 100 Ω-µm2

GMR (∆R/R) ~ < 50-90+%(varies with barrier thickness)

Critical current densities quite similar in good spin valves and MTJsHigh polarization of MTJs may give a ~ 2x advantage

Nanopillar MAGNETIC TUNNEL JUNCTION

Py (2 nm)

Py (12 nm)AlOx (~0.7 nm

Cu

Cu

free layer

fixed layer

Practical issues for spin-torque switching: speed, switching currents, impedance

5.1

5.3

5.5

0 600 1200Magnetic Field [G]

dV/d

I [O

hm]

5.1

5.3

5.5

-1 -0.5 0 0.5 1Current [mA]

dV/d

I [O

hm]

T = 4.2 KNanopillar Spin-Valve

Py (2 nm)

Py (12 nm)Cu (6 nm)

Cu

Cu

free layer

fixed layer

Spin Transfer Driven Magnetic Reversal

~120 nm

~40 nm

ChallengesIn “standard” nanopillar devices, initial direction of spin torque is determined by a random thermal fluctuation from equilibrium. This leads to a random phase of the precessional dynamics.

Time-resolved measurements require devices with a non-zero angle between the free and the fixed layers.

( )θτ sin~ 2 ⋅=××→→→→

ImmImst

fixed layer

M

free layer

m

τst

free layer

m

fixed layer

M

τst

1)sin( <<θ 1)sin( ≈θ

V(t) < 1 mV, ∆t ~ 10-100 psec

Time Resolved Measurements of Nanomagnet Dynamics

T > 0

Sampling OscilloscopeStep Generator

dc

+25 dB

Measurements of Spin-Transfer Dynamics

Py (4 nm)

Py (4 nm)Cu (8 nm)

Cu

Cu

free layer

fixed layerIrMn (8 nm)

~ 130 nm

~ 60 nmHEB

HEB = exchange bias field

I. N. Krivorotov et al.Science 307, 228 (2005).

Exchange biasing of the fixed Py layer at 45º to the easy axis results in a non-zero initial angle between magnetic moments of the fixed and free layers. This establishes a well-defined phase for precessional dynamics of the magnet.

5.9

5.95

6

6.05

6.1

-400 -200 0 200 400 600

Filed (G)

dV/d

I (O

hm)

Happlied

Mfixed

Mfree

θ0

- data- Stoner-Wolfarth fit

Equilibrium Configuration of Magnetization

θ0~ 35°

( )( )2/cos1

2/cos12

2

0 θχθ

+−

∆+= RRR

χ = 0.5; Heb = 1.5 kG

Sample 2

High Speed Spin Torque Switching

switching time1 → τ =

θ0 ~ initial angle betweenmagnetizations

-set by thermal fluctuations or magnetic pinning

Ic0 is T= 0 critical current

co

0

II2

πln

−

⎟⎠⎞⎜

⎝⎛

θ

1J.Sun, Phys Rev B. 62, 570 (2000)

Faster reversal requires larger Iswitch

Spin polarized current must deliver sufficient spin angular momentum to nanomagnet to reverse magnetic moment.

Hence (I –Ic0)x τ = constant

How fast is spin-transfer-driven switching?

Sampling Oscilloscope

Step Generator

dc

+25 dB

Switching time < 1 ns at high pulse amplitude

Measure time dependent response of nanopillar resistance to step pulse.

I. N. Krivorotov et al.Science 307, 228 (2005).

Ico+ = α e Ms Vol [H + Han + 2 π Ms ] / hg(0) ≈ 2 π α e Ms

2 Vol / h g(0)

Ico- = α e Ms Vol [H - Han - 2 π Ms ] / hg(π) ≈ 2 π α e Ms

2 Vol / h g(π)

Jco+ ≈ 2 π α e Ms

2 t / h g(0); Jco+ ≈ 2 π α e Ms

2 t / h g(π)

t = nanomagnet thickness, α =Gilbert damping parameter, Ms = magnetization

Han = shape anisotropy field

Critical Current for Spin Torque Switching

Han

4 π Ms

out of plane demagnetization

field top view

To reduce Jco - reduce t, Ms and/or α but must maintain nanomagnet stability

This requires UK = MsHan Vol /2 > 50 kBT - ten year bit stability

Ic ∝ Ms2 α (Vol)

U0 ∝ HanMs(Vol)Han ~ Ms(t/t0)

U0

MRAM requirement:Bit lifetime ~ 10 years → U0 = 1 eV at RTWith heating to 100º C → U0 = 1.3 eV

~120 nm

~40 nm

Minimize Ms and sample volume Use shape anisotropy to maximize Hk

thick and elongated

Decreasing Switching Currents

4.5 nm Py : U0,P-AP=0.85 eV, Ic0+ = .42 mA

U0,AP-P=0.73 eV, Ic0- = .39 mA

Ic0 = zero-temp critical current. Need Ico < 100 µANeed to decrease damping and improve micromagnetics

Spin torque switching currents of low Ms free layers

Pulse-response measurementsPulse Generator

dc

+25 dB

Apply current pulse to device.

Determine if pulse has switched device.

Increase pulse duration until probability of switching goes to unity.

Increase current pulse amplitude and repeat.

0.0 0.5 1.0 1.5 2.0 2.5

0.0

0.2

0.4

0.6

0.8

1.0

Sw

itchi

ng P

roba

bilit

y

Pulse Amplitude (mA)

100 ns30 ns10 ns 3 ns1 ns

Comparison with Single Domain LLG Simulations

0.0

0.2

0.4

0.6

0.8

1.0

Sw

itchi

ng P

roba

bilit

y

0.5 1.0 1.5 2.0 2.5

simulationsdata

Pulse Amplitude (mA)

1 ns3 ns100 ns

0.0 0.2 0.4 0.6 0.8 1.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

p-apap-p

I50

%(m

A)

τ -1 (ns -1 )

Fitting to LLG simulation yields empirical spin-torque function and damping

N.B. Similar AP-P and P-AP switching currents in these devicesBraganca et al. APL ‘05

Spin Transfer Torque Function

– effect of device geometry on g(θ )–spin accumulation affects?

0 π /2 π

0.050.1

0.150.2

0.250.3

g

g(θ) – Slonczewski 1996

g(θ) – Cornell (exp.)

Ic, P - AP ~ g’(0) ; Ic, AP - A ~ g’(π)

( ) ( )mImmmHmmeff ××+⎟

⎠⎞

⎜⎝⎛ ×−×=

)sin()(

2 θθµαγ

mg

edtd

dtd B

g(θ) – Xiao, et al.

See also: Manschot et al., APL.2004Barnas et al. PRB 2005

)cos(1)sin()(θ

θθB

Ag+

=

>> λsf> λsf

Effect of Electrode Structure on Spin Torque

Net electron flow

0.050.1

0.150.2

0.250.3

gold cap

g

FeM

n

>> λsf> λsf

Effect of Electrode Structure on Spin Torque

Net electron flow

0.050.1

0.150.2

0.250.3

gold cap

Fe-Mn cap

g

Pulsed Current ExperimentsPt Capped Devices

1 2 3 4 50.0

0.2

0.4

0.6

0.8

1.0

1 ns pulse data 3 ns pulse data 10 ns pulse data 100 ns pulse datasimulations

Sw

itchi

ng P

roba

bilit

y

Pulse Amplitude (mA)

A = 0.18α = 0.037

Standard Configuration

1 2 3 4 5

1 ns pulse data 3 ns pulse data 10 ns pulse data 100 ns pulse datasimulations

Pulse Amplitude (mA)

A = 0.52α = 0.047

Inverted Configuration

AP-P switching

Spin pumping enhancement in inverted samples → Better spin sinking in extended Cu lead

LLG fit deviation from data at large currents – microwave oscillations

)cos(1)sin()(θ

θθB

Ag+

=Torque angular dependence

A – Torque amplitude – from spin current and spin accumulation

LLG simulations

γγ

+−

=11B

APPswitch

PAPswitch

II

→

→=,

,γ

e-

0.08-0.130.11-0.230.32-0.330.02-0.19B

0.0470.033-0.0370.033-0.0370.025-0.030α

0.45-0.520.18-0.210.12-0.160.25-0.30A

Pt inv.Pt capFe-Mncap Au cap

Pt normal Pt inverted

A=0.18B=0.23

A=0.52B=0.13

30nm hole

150x250nm pillar

Pt 30nm

Cu

Py 20nm

Cu 8nm

Py 5nm

Cu

Spin-Transfer-Switching by Spatially Non-Uniform Currents

Al2O3 3nmSiO2 SiO2

15-30nm aperture sizes

150nm

A 3nm Al2O3 insulating barrier with a nano-orifice is inserted into a Cu/Py spin-valve nanopillarGoal:

Result:

150x250 nm pillar

Hc~37.5Oe ∆R~253mΩ

150mΩ

J ~ 1.2x107 A/cm2AP-PIc- = 4 mA

P-APIc+ = 7.8 mA

100 x 200 nm2

uniform current

Jpillar ~ 4x105 A/cm2

Jhole~1.6x107 A/cm2

AP-PIc- = 50 µΑ

P-APIc+ = 180 µA

150 x 250 nm2

with 30 nm aperture

T=4.2K

R = 3 Ω

R = 12 Ω

•The nano-aperture device requires much less current to induce switching than a nanopillar with uniform current flow.•Current-induced switching may not result in full reversal of the nanomagnet

11.65

11.7

11.75

11.8

11.85

11.9

11.95

-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5I(mA)

dV/d

I( Ω)

11.6511.7

11.7511.8

11.8511.9

11.9512

-600 -400 -200 0

H(Oe)

R( Ω

)

150mΩ

T=4.2K

Spin-Transfer-Switching by Spatially Non-Uniform Currents

-1 ´10 - 7

-5 ´10 - 8

05´10 - 8

1´10 - 7

-5 ´10 - 8

0

5´10 - 8

02´10 6

4´10 6

6´10 6

10 - 7

-5 ´10 - 8

05´10 - 8

´ - 7

3D OOMMF Simulations

The effect of spin torque was modeled using LLG equation with the Slonczweskiterm for each cell. The simulations were performed taking into account the Oersted field created by electron flow through a wire.

OOMMF is a public

software developed by M.J.Donahue

and D.G. Porter from

NIST

t=1.13ns t=1.6ns

t=2.3ns

t=3.3ns

t=2.06ns

t=2.5ns

t=3.96ns t=5.9ns

0.5 mA

Spin Transfer with Magnetic Tunnel Junctions

0.1 1 10 100 100010000

Bad TMR,

Pinholes

Good TMR, too high resistance to do spin transfer.

RA (Ω µm2)

Ok

for s

pin

tran

sfer

Pt 30 nm

Cu 5 nm

CoFeB 2 nmAlOx 7-8 Å

CoFeB 8 nm

Cu 80 nm

Ta 10 nm 147 nm

56 nm

147 nm

56 nm

Challenge: Tunnel barriers with high TMR that can withstand the currents necessary for switching, particularly for fast switching

Early Demonstrations with AlOx

Minor LoopH = 387 OeT= 77K

• There is a small TMR measured with DC resistance at switching currents.

• Wear-out of barriers a concern due to high critical currents/voltages

T = 77 K

Switching currents

Huai et. al., APL 84, 3118 (2004)

Fuchs et. al., APL 85, 1205 (2004)

20 Å CoFeB

80 Å CoFeB6.5 Å Al + Oxygen

CoFeB=Co88.2Fe9.8B2

Anti-aligned fixed layers

Aligned fixed layers

Spins from each fixed layer are in the same direction – more spin torque

Spins from each fixed layer are in opposite directions – almost no spin torque

5 nm CoFe

6 nm Cu

4 nm Py~0.8 nm AlOx8 nm CoFe

20 nm Ta

Increasing spin torque in MTJs with three magnetic layers

P

APAP/P

AP/P

APAP

PPT=77KAnti-aligned fixed layers Aligned fixed layers

Ic,o+ = 0.29±0.01 mAIc,o- = -0.28±0.01 mA

Jc,o/t = (2.9 ± 0.4) x106 A/(cm2-nm), reduced by 40% compared to a Py free layer with one fixed layer: 5x106 A/(cm2-nm)

(shape and size not optimized)

G. D. Fuchs et al., Appl. Phys. Lett. 86, 152509 (2005).

Ohmic heating reduces Hc,minimal spin torque

Strong spin torque

Spin Transfer Switching in 3-layer MTJs

Note the similarity of Ic’s

Questions regarding spin torque in MTJs

• Why does TMR decrease with increasing bias?

• How does bias affect spin-transfer torque?

• What is the nature of spin polarized transport in MgObased MTJs at finite bias?

Models that describe TMR(V) must also be consistent with spin torque, Nst/I(I) and I(V)

-0.3V 0 0.3V

How to measure torque vs. current

A thermally stable free layer can only provide a measure of the spin-torque at the switching bias

A thermally unstable free layer can provide a measure ofspin-torque continuously as a function of bias by applying H and I so as to have opposing effects

Sample structure

Lacour et al, APL 85, 4681, (2004)

• Bottom pinned SAF nearly cancels the dipole field and has a very large exchange field (~2 kOe)

• Devices are patterned with a 2:1 aspect ratio

• Have a range of thermal activation barriers

CoFe = Co86Fe14

Py = Ni91.5Fe8.5

CoFe 1 nm/Py 1.8 nmMgO 0.8 nmCoFe 1.9 nmRu 0.7 nmCoFe 2.2 nm

PtMn 15.4 nm

100 nm

Katine and Mauri - HGST

Experimental approach

⎥⎥

⎦

⎤

⎢⎢

⎣

⎡

⎟⎟⎠

⎞⎜⎜⎝

⎛⎟⎟⎠

⎞⎜⎜⎝

⎛ −±=

ococ

dip

B

aoAPP I

IIH

HHTk

EExp

,

2

,/

)(11 γττ mLifetime in thermal activation regime

γ(I)=Scaling factor to parameterize Nst/Ivariation with I - “Spin Transfer Efficiency”

E. B. Myers, et al, PRL 89, 196801 (2002).Z. Li and S. Zhang, PRB 68, 024404 (2003).I. N. Krivorotov, et al, PRL 93 166603 (2004).

Positions of equal mean lifetimes if the efficiency is constant with bias

Positions of equal mean lifetimes if efficiency decreases with increasing bias

Increasing Current (I)

Mag

netic

Fie

ld (H

)

0

H(I) data - Linear Response

TMR decreases by over 40%

Hd

H(I) data - Linear Response

TMR decreases by over 40%

Break in data – crystalline anisotropy effect

Spin Transfer Efficiency•Data are consistent with less than a 10% decrease in spin torqueefficiency out to the switching bias point (~ 0.3 V)

Tunnel Conductance Through MgO“s-like”

“pd-like”No s-like channels!

s-like decays in the electrode

No s-like channels!

W. H. Butler, X. –G. Zhang, T. C. Schulthess, PRB 63, 054416 (2001).J. Mathon and A. Umerski, PRB 63, 220403 (2001).

MgO DOS Data

-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6

Fe / 2nm MgO[eb] Fe / 2nm MgO[rf] CoFeB / 2nm MgO[rf] 375C 1 hr

DO

S (A

.U.)

Negative Tip Bias (V)

Negative Tip Bias (V)

DO

S (A

.U.)

Fe / 20Å MgO[eb]Fe / 20Å MgO[rf]CoFeB / 20Å MgO[rf] 375oC 1hr

5.5 eV

2 eV

STM tunneling spectroscopy evidence for O vacancy defects in MgO barrier layers

Tunnel Conductance through MgO

Simmon’s model fit:

=1.35±0.05

=0.82±0.02 *pm

*apm

Magnetic state dependent effective mass (decay length):

Elastic scattering by barrier defects reduces the TMR

P

AP

γ(I)~const implies that:

• conductance for each spin channel varies with bias at a rate proportional to the zero bias DOS.

•electron scattering rate from defects is not strongly spin dependent!

W. H. Butler, X. –G. Zhang, T. C. Schulthess, PRB 63, 054416 (2001).J. Mathon and A. Umerski, PRB 63, 220403 (2001).

Symmetry of Critical Currents

])(1[2)()( 2 θ

θCosVP

VPg+

=

Polarization term

Asymmetry term is present to convert Slonczewski’s critical voltage (Vc) into a critical current (Jc).

A better approximation:

⎥⎦

⎤⎢⎣

⎡⎟⎟⎠

⎞⎜⎜⎝

⎛+

+

==

θθ

CosVTMR

VTMRVPg

)(2)(12

)0()(

P2 calculated from TMR(V)

Polarization term is a constant function of V, consistent with our study

Diao et al., APL 87, 232502, (2005)

Conclusions – ST in MTJs

•Spin-transfer torque per unit current is independent of bias within 10% up to 0.35 V (good news for spin-torque driven MRAM)

•Measurement brings new information to help understand the relationship between bias and spin-polarized tunneling

•Results are inconsistent with:

Free-electron, split-band tunneling models

Magnon emission models that reduce polarization factors

•Results are consistent with calculations due to Butler et al and Mathonet al for transport through ultra-thin MgO tunnel barriers allowing for defects in non-ideal tunnel barriers.

Fuchs et al., cond-mat/0510786